Research agents

The shape

A research run is usually one prompt — “survey the state of X, cite sources, give me a 1-page brief” — handed to an assistant with research-flavored skills enabled. The agent uses the browser, summarizes findings, and writes the result back into your workspace.

Skills that pair well

• `browser-automation` — open, search, scroll, screenshot.
• `summarize` — condense long pages and transcripts.
• `openai-whisper` — transcribe video or podcast sources before summarizing.
• `obsidian` — if your vault is your second brain, drop the brief there with backlinks.

What ends up in memory

Every brief the agent writes lands in workspace/memory/ or workspace/kb/ as a markdown file. The indexer chunks it and adds it to the vector store, so the next research run can pull the previous findings as context automatically.

Schedule it

Pair this with a routine to wake the agent up weekly and produce a new edition of the brief. The schedule type is Weekly; the prompt is the same one you’d type by hand.